What I'm looking out for is a problem definition to work on multi-agent systems preferably for task allocation. The state or health of each individual agent could be analysed & the task would be allocated as per that condition. In case there are observations about the 'health' of any agent failing a preventive algorithmic approach could relieve the agent of that task by re-distributing it to other agents.